]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/fs/nfs/read.c | |
3 | * | |
4 | * Block I/O for NFS | |
5 | * | |
6 | * Partial copy of Linus' read cache modifications to fs/nfs/file.c | |
7 | * modified for async RPC by okir@monad.swb.de | |
1da177e4 LT |
8 | */ |
9 | ||
1da177e4 LT |
10 | #include <linux/time.h> |
11 | #include <linux/kernel.h> | |
12 | #include <linux/errno.h> | |
13 | #include <linux/fcntl.h> | |
14 | #include <linux/stat.h> | |
15 | #include <linux/mm.h> | |
16 | #include <linux/slab.h> | |
17 | #include <linux/pagemap.h> | |
18 | #include <linux/sunrpc/clnt.h> | |
19 | #include <linux/nfs_fs.h> | |
20 | #include <linux/nfs_page.h> | |
64419a9b AA |
21 | #include <linux/smp_lock.h> |
22 | #include <linux/module.h> | |
1da177e4 LT |
23 | |
24 | #include <asm/system.h> | |
bae724ef | 25 | #include "pnfs.h" |
1da177e4 | 26 | |
f11c88af | 27 | #include "nfs4_fs.h" |
49a70f27 | 28 | #include "internal.h" |
91d5b470 | 29 | #include "iostat.h" |
9a9fc1c0 | 30 | #include "fscache.h" |
91d5b470 | 31 | |
1da177e4 LT |
32 | #define NFSDBG_FACILITY NFSDBG_PAGECACHE |
33 | ||
bae724ef FI |
34 | static int nfs_pagein_multi(struct inode *, struct list_head *, unsigned int, size_t, int, struct pnfs_layout_segment *); |
35 | static int nfs_pagein_one(struct inode *, struct list_head *, unsigned int, size_t, int, struct pnfs_layout_segment *); | |
ec06c096 TM |
36 | static const struct rpc_call_ops nfs_read_partial_ops; |
37 | static const struct rpc_call_ops nfs_read_full_ops; | |
1da177e4 | 38 | |
e18b890b | 39 | static struct kmem_cache *nfs_rdata_cachep; |
3feb2d49 | 40 | static mempool_t *nfs_rdata_mempool; |
1da177e4 LT |
41 | |
42 | #define MIN_POOL_READ (32) | |
43 | ||
8d5658c9 | 44 | struct nfs_read_data *nfs_readdata_alloc(unsigned int pagecount) |
3feb2d49 | 45 | { |
93870d76 | 46 | struct nfs_read_data *p = mempool_alloc(nfs_rdata_mempool, GFP_KERNEL); |
3feb2d49 TM |
47 | |
48 | if (p) { | |
49 | memset(p, 0, sizeof(*p)); | |
50 | INIT_LIST_HEAD(&p->pages); | |
e9f7bee1 | 51 | p->npages = pagecount; |
0d0b5cb3 CL |
52 | if (pagecount <= ARRAY_SIZE(p->page_array)) |
53 | p->pagevec = p->page_array; | |
3feb2d49 | 54 | else { |
93870d76 | 55 | p->pagevec = kcalloc(pagecount, sizeof(struct page *), GFP_KERNEL); |
0d0b5cb3 | 56 | if (!p->pagevec) { |
3feb2d49 TM |
57 | mempool_free(p, nfs_rdata_mempool); |
58 | p = NULL; | |
59 | } | |
60 | } | |
61 | } | |
62 | return p; | |
63 | } | |
64 | ||
1ae88b2e | 65 | void nfs_readdata_free(struct nfs_read_data *p) |
3feb2d49 TM |
66 | { |
67 | if (p && (p->pagevec != &p->page_array[0])) | |
68 | kfree(p->pagevec); | |
69 | mempool_free(p, nfs_rdata_mempool); | |
70 | } | |
71 | ||
1ae88b2e | 72 | static void nfs_readdata_release(struct nfs_read_data *rdata) |
1da177e4 | 73 | { |
bae724ef | 74 | put_lseg(rdata->lseg); |
383ba719 TM |
75 | put_nfs_open_context(rdata->args.context); |
76 | nfs_readdata_free(rdata); | |
1da177e4 LT |
77 | } |
78 | ||
1da177e4 LT |
79 | static |
80 | int nfs_return_empty_page(struct page *page) | |
81 | { | |
eebd2aa3 | 82 | zero_user(page, 0, PAGE_CACHE_SIZE); |
1da177e4 LT |
83 | SetPageUptodate(page); |
84 | unlock_page(page); | |
85 | return 0; | |
86 | } | |
87 | ||
1de3fc12 TM |
88 | static void nfs_readpage_truncate_uninitialised_page(struct nfs_read_data *data) |
89 | { | |
90 | unsigned int remainder = data->args.count - data->res.count; | |
91 | unsigned int base = data->args.pgbase + data->res.count; | |
92 | unsigned int pglen; | |
93 | struct page **pages; | |
94 | ||
95 | if (data->res.eof == 0 || remainder == 0) | |
96 | return; | |
97 | /* | |
98 | * Note: "remainder" can never be negative, since we check for | |
99 | * this in the XDR code. | |
100 | */ | |
101 | pages = &data->args.pages[base >> PAGE_CACHE_SHIFT]; | |
102 | base &= ~PAGE_CACHE_MASK; | |
103 | pglen = PAGE_CACHE_SIZE - base; | |
79558f36 TM |
104 | for (;;) { |
105 | if (remainder <= pglen) { | |
eebd2aa3 | 106 | zero_user(*pages, base, remainder); |
79558f36 TM |
107 | break; |
108 | } | |
eebd2aa3 | 109 | zero_user(*pages, base, pglen); |
79558f36 TM |
110 | pages++; |
111 | remainder -= pglen; | |
112 | pglen = PAGE_CACHE_SIZE; | |
113 | base = 0; | |
114 | } | |
1de3fc12 TM |
115 | } |
116 | ||
f42b293d DH |
117 | int nfs_readpage_async(struct nfs_open_context *ctx, struct inode *inode, |
118 | struct page *page) | |
1da177e4 LT |
119 | { |
120 | LIST_HEAD(one_request); | |
121 | struct nfs_page *new; | |
122 | unsigned int len; | |
123 | ||
49a70f27 | 124 | len = nfs_page_length(page); |
1da177e4 LT |
125 | if (len == 0) |
126 | return nfs_return_empty_page(page); | |
127 | new = nfs_create_request(ctx, inode, page, 0, len); | |
128 | if (IS_ERR(new)) { | |
129 | unlock_page(page); | |
130 | return PTR_ERR(new); | |
131 | } | |
132 | if (len < PAGE_CACHE_SIZE) | |
eebd2aa3 | 133 | zero_user_segment(page, len, PAGE_CACHE_SIZE); |
1da177e4 | 134 | |
1da177e4 | 135 | nfs_list_add_request(new, &one_request); |
bcb71bba | 136 | if (NFS_SERVER(inode)->rsize < PAGE_CACHE_SIZE) |
bae724ef | 137 | nfs_pagein_multi(inode, &one_request, 1, len, 0, NULL); |
bcb71bba | 138 | else |
bae724ef | 139 | nfs_pagein_one(inode, &one_request, 1, len, 0, NULL); |
1da177e4 LT |
140 | return 0; |
141 | } | |
142 | ||
143 | static void nfs_readpage_release(struct nfs_page *req) | |
144 | { | |
7f8e05f6 DH |
145 | struct inode *d_inode = req->wb_context->path.dentry->d_inode; |
146 | ||
147 | if (PageUptodate(req->wb_page)) | |
148 | nfs_readpage_to_fscache(d_inode, req->wb_page, 0); | |
149 | ||
1da177e4 LT |
150 | unlock_page(req->wb_page); |
151 | ||
1da177e4 | 152 | dprintk("NFS: read done (%s/%Ld %d@%Ld)\n", |
88be9f99 TM |
153 | req->wb_context->path.dentry->d_inode->i_sb->s_id, |
154 | (long long)NFS_FILEID(req->wb_context->path.dentry->d_inode), | |
1da177e4 LT |
155 | req->wb_bytes, |
156 | (long long)req_offset(req)); | |
10d2c46f | 157 | nfs_release_request(req); |
1da177e4 LT |
158 | } |
159 | ||
dc70d7b3 | 160 | int nfs_initiate_read(struct nfs_read_data *data, struct rpc_clnt *clnt, |
64419a9b | 161 | const struct rpc_call_ops *call_ops) |
1da177e4 | 162 | { |
64419a9b | 163 | struct inode *inode = data->inode; |
84115e1c | 164 | int swap_flags = IS_SWAPFILE(inode) ? NFS_RPC_SWAPFLAGS : 0; |
07737691 | 165 | struct rpc_task *task; |
bdc7f021 TM |
166 | struct rpc_message msg = { |
167 | .rpc_argp = &data->args, | |
168 | .rpc_resp = &data->res, | |
64419a9b | 169 | .rpc_cred = data->cred, |
bdc7f021 | 170 | }; |
84115e1c | 171 | struct rpc_task_setup task_setup_data = { |
07737691 | 172 | .task = &data->task, |
64419a9b | 173 | .rpc_client = clnt, |
bdc7f021 | 174 | .rpc_message = &msg, |
84115e1c TM |
175 | .callback_ops = call_ops, |
176 | .callback_data = data, | |
101070ca | 177 | .workqueue = nfsiod_workqueue, |
84115e1c TM |
178 | .flags = RPC_TASK_ASYNC | swap_flags, |
179 | }; | |
1da177e4 | 180 | |
64419a9b AA |
181 | /* Set up the initial task struct. */ |
182 | NFS_PROTO(inode)->read_setup(data, &msg); | |
183 | ||
184 | dprintk("NFS: %5u initiated read call (req %s/%lld, %u bytes @ " | |
185 | "offset %llu)\n", | |
186 | data->task.tk_pid, | |
187 | inode->i_sb->s_id, | |
188 | (long long)NFS_FILEID(inode), | |
189 | data->args.count, | |
190 | (unsigned long long)data->args.offset); | |
191 | ||
192 | task = rpc_run_task(&task_setup_data); | |
193 | if (IS_ERR(task)) | |
194 | return PTR_ERR(task); | |
195 | rpc_put_task(task); | |
196 | return 0; | |
197 | } | |
dc70d7b3 | 198 | EXPORT_SYMBOL_GPL(nfs_initiate_read); |
64419a9b AA |
199 | |
200 | /* | |
201 | * Set up the NFS read request struct | |
202 | */ | |
203 | static int nfs_read_rpcsetup(struct nfs_page *req, struct nfs_read_data *data, | |
204 | const struct rpc_call_ops *call_ops, | |
205 | unsigned int count, unsigned int offset, | |
206 | struct pnfs_layout_segment *lseg) | |
207 | { | |
208 | struct inode *inode = req->wb_context->path.dentry->d_inode; | |
209 | ||
1da177e4 | 210 | data->req = req; |
84115e1c | 211 | data->inode = inode; |
64419a9b | 212 | data->cred = req->wb_context->cred; |
bae724ef | 213 | data->lseg = get_lseg(lseg); |
1da177e4 LT |
214 | |
215 | data->args.fh = NFS_FH(inode); | |
216 | data->args.offset = req_offset(req) + offset; | |
217 | data->args.pgbase = req->wb_pgbase + offset; | |
218 | data->args.pages = data->pagevec; | |
219 | data->args.count = count; | |
383ba719 | 220 | data->args.context = get_nfs_open_context(req->wb_context); |
f11ac8db | 221 | data->args.lock_context = req->wb_lock_context; |
1da177e4 LT |
222 | |
223 | data->res.fattr = &data->fattr; | |
224 | data->res.count = count; | |
225 | data->res.eof = 0; | |
0e574af1 | 226 | nfs_fattr_init(&data->fattr); |
1da177e4 | 227 | |
64419a9b AA |
228 | if (data->lseg && |
229 | (pnfs_try_to_read_data(data, call_ops) == PNFS_ATTEMPTED)) | |
230 | return 0; | |
bdc7f021 | 231 | |
64419a9b | 232 | return nfs_initiate_read(data, NFS_CLIENT(inode), call_ops); |
1da177e4 LT |
233 | } |
234 | ||
235 | static void | |
236 | nfs_async_read_error(struct list_head *head) | |
237 | { | |
238 | struct nfs_page *req; | |
239 | ||
240 | while (!list_empty(head)) { | |
241 | req = nfs_list_entry(head->next); | |
242 | nfs_list_remove_request(req); | |
243 | SetPageError(req->wb_page); | |
244 | nfs_readpage_release(req); | |
245 | } | |
246 | } | |
247 | ||
1da177e4 LT |
248 | /* |
249 | * Generate multiple requests to fill a single page. | |
250 | * | |
251 | * We optimize to reduce the number of read operations on the wire. If we | |
252 | * detect that we're reading a page, or an area of a page, that is past the | |
253 | * end of file, we do not generate NFS read operations but just clear the | |
254 | * parts of the page that would have come back zero from the server anyway. | |
255 | * | |
256 | * We rely on the cached value of i_size to make this determination; another | |
257 | * client can fill pages on the server past our cached end-of-file, but we | |
258 | * won't see the new data until our attribute cache is updated. This is more | |
259 | * or less conventional NFS client behavior. | |
260 | */ | |
bae724ef | 261 | static int nfs_pagein_multi(struct inode *inode, struct list_head *head, unsigned int npages, size_t count, int flags, struct pnfs_layout_segment *lseg) |
1da177e4 LT |
262 | { |
263 | struct nfs_page *req = nfs_list_entry(head->next); | |
264 | struct page *page = req->wb_page; | |
265 | struct nfs_read_data *data; | |
e9f7bee1 TM |
266 | size_t rsize = NFS_SERVER(inode)->rsize, nbytes; |
267 | unsigned int offset; | |
1da177e4 | 268 | int requests = 0; |
dbae4c73 | 269 | int ret = 0; |
1da177e4 LT |
270 | LIST_HEAD(list); |
271 | ||
272 | nfs_list_remove_request(req); | |
273 | ||
bcb71bba | 274 | nbytes = count; |
e9f7bee1 TM |
275 | do { |
276 | size_t len = min(nbytes,rsize); | |
277 | ||
8d5658c9 | 278 | data = nfs_readdata_alloc(1); |
1da177e4 LT |
279 | if (!data) |
280 | goto out_bad; | |
1da177e4 LT |
281 | list_add(&data->pages, &list); |
282 | requests++; | |
e9f7bee1 TM |
283 | nbytes -= len; |
284 | } while(nbytes != 0); | |
1da177e4 LT |
285 | atomic_set(&req->wb_complete, requests); |
286 | ||
bae724ef FI |
287 | /* We know lseg==NULL */ |
288 | lseg = pnfs_update_layout(inode, req->wb_context, IOMODE_READ); | |
1da177e4 LT |
289 | ClearPageError(page); |
290 | offset = 0; | |
bcb71bba | 291 | nbytes = count; |
1da177e4 | 292 | do { |
dbae4c73 TM |
293 | int ret2; |
294 | ||
1da177e4 LT |
295 | data = list_entry(list.next, struct nfs_read_data, pages); |
296 | list_del_init(&data->pages); | |
297 | ||
298 | data->pagevec[0] = page; | |
1da177e4 | 299 | |
bcb71bba TM |
300 | if (nbytes < rsize) |
301 | rsize = nbytes; | |
dbae4c73 | 302 | ret2 = nfs_read_rpcsetup(req, data, &nfs_read_partial_ops, |
bae724ef | 303 | rsize, offset, lseg); |
dbae4c73 TM |
304 | if (ret == 0) |
305 | ret = ret2; | |
bcb71bba TM |
306 | offset += rsize; |
307 | nbytes -= rsize; | |
1da177e4 | 308 | } while (nbytes != 0); |
bae724ef | 309 | put_lseg(lseg); |
1da177e4 | 310 | |
dbae4c73 | 311 | return ret; |
1da177e4 LT |
312 | |
313 | out_bad: | |
314 | while (!list_empty(&list)) { | |
315 | data = list_entry(list.next, struct nfs_read_data, pages); | |
316 | list_del(&data->pages); | |
317 | nfs_readdata_free(data); | |
318 | } | |
319 | SetPageError(page); | |
320 | nfs_readpage_release(req); | |
321 | return -ENOMEM; | |
322 | } | |
323 | ||
bae724ef | 324 | static int nfs_pagein_one(struct inode *inode, struct list_head *head, unsigned int npages, size_t count, int flags, struct pnfs_layout_segment *lseg) |
1da177e4 LT |
325 | { |
326 | struct nfs_page *req; | |
327 | struct page **pages; | |
328 | struct nfs_read_data *data; | |
dbae4c73 | 329 | int ret = -ENOMEM; |
1da177e4 | 330 | |
8d5658c9 | 331 | data = nfs_readdata_alloc(npages); |
bae724ef FI |
332 | if (!data) { |
333 | nfs_async_read_error(head); | |
334 | goto out; | |
335 | } | |
1da177e4 | 336 | |
1da177e4 | 337 | pages = data->pagevec; |
1da177e4 LT |
338 | while (!list_empty(head)) { |
339 | req = nfs_list_entry(head->next); | |
340 | nfs_list_remove_request(req); | |
341 | nfs_list_add_request(req, &data->pages); | |
342 | ClearPageError(req->wb_page); | |
343 | *pages++ = req->wb_page; | |
1da177e4 LT |
344 | } |
345 | req = nfs_list_entry(data->pages.next); | |
bae724ef FI |
346 | if ((!lseg) && list_is_singular(&data->pages)) |
347 | lseg = pnfs_update_layout(inode, req->wb_context, IOMODE_READ); | |
1da177e4 | 348 | |
bae724ef FI |
349 | ret = nfs_read_rpcsetup(req, data, &nfs_read_full_ops, count, 0, lseg); |
350 | out: | |
351 | put_lseg(lseg); | |
dbae4c73 | 352 | return ret; |
1da177e4 LT |
353 | } |
354 | ||
0b671301 TM |
355 | /* |
356 | * This is the callback from RPC telling us whether a reply was | |
357 | * received or some error occurred (timeout or socket shutdown). | |
358 | */ | |
359 | int nfs_readpage_result(struct rpc_task *task, struct nfs_read_data *data) | |
360 | { | |
361 | int status; | |
362 | ||
3110ff80 | 363 | dprintk("NFS: %s: %5u, (status %d)\n", __func__, task->tk_pid, |
0b671301 TM |
364 | task->tk_status); |
365 | ||
366 | status = NFS_PROTO(data->inode)->read_done(task, data); | |
367 | if (status != 0) | |
368 | return status; | |
369 | ||
370 | nfs_add_stats(data->inode, NFSIOS_SERVERREADBYTES, data->res.count); | |
371 | ||
372 | if (task->tk_status == -ESTALE) { | |
3a10c30a | 373 | set_bit(NFS_INO_STALE, &NFS_I(data->inode)->flags); |
0b671301 TM |
374 | nfs_mark_for_revalidate(data->inode); |
375 | } | |
0b671301 TM |
376 | return 0; |
377 | } | |
378 | ||
fdd1e74c | 379 | static void nfs_readpage_retry(struct rpc_task *task, struct nfs_read_data *data) |
0b671301 TM |
380 | { |
381 | struct nfs_readargs *argp = &data->args; | |
382 | struct nfs_readres *resp = &data->res; | |
383 | ||
384 | if (resp->eof || resp->count == argp->count) | |
d61e612a | 385 | return; |
0b671301 TM |
386 | |
387 | /* This is a short read! */ | |
388 | nfs_inc_stats(data->inode, NFSIOS_SHORTREAD); | |
389 | /* Has the server at least made some progress? */ | |
390 | if (resp->count == 0) | |
d61e612a | 391 | return; |
0b671301 TM |
392 | |
393 | /* Yes, so retry the read at the end of the data */ | |
394 | argp->offset += resp->count; | |
395 | argp->pgbase += resp->count; | |
396 | argp->count -= resp->count; | |
0110ee15 | 397 | nfs_restart_rpc(task, NFS_SERVER(data->inode)->nfs_client); |
0b671301 TM |
398 | } |
399 | ||
1da177e4 LT |
400 | /* |
401 | * Handle a read reply that fills part of a page. | |
402 | */ | |
ec06c096 | 403 | static void nfs_readpage_result_partial(struct rpc_task *task, void *calldata) |
1da177e4 | 404 | { |
ec06c096 | 405 | struct nfs_read_data *data = calldata; |
1da177e4 | 406 | |
ec06c096 TM |
407 | if (nfs_readpage_result(task, data) != 0) |
408 | return; | |
fdd1e74c TM |
409 | if (task->tk_status < 0) |
410 | return; | |
0b671301 | 411 | |
fdd1e74c TM |
412 | nfs_readpage_truncate_uninitialised_page(data); |
413 | nfs_readpage_retry(task, data); | |
414 | } | |
415 | ||
416 | static void nfs_readpage_release_partial(void *calldata) | |
417 | { | |
418 | struct nfs_read_data *data = calldata; | |
419 | struct nfs_page *req = data->req; | |
420 | struct page *page = req->wb_page; | |
421 | int status = data->task.tk_status; | |
422 | ||
423 | if (status < 0) | |
0b671301 | 424 | SetPageError(page); |
fdd1e74c | 425 | |
1da177e4 LT |
426 | if (atomic_dec_and_test(&req->wb_complete)) { |
427 | if (!PageError(page)) | |
428 | SetPageUptodate(page); | |
429 | nfs_readpage_release(req); | |
430 | } | |
fdd1e74c | 431 | nfs_readdata_release(calldata); |
1da177e4 LT |
432 | } |
433 | ||
f11c88af AA |
434 | #if defined(CONFIG_NFS_V4_1) |
435 | void nfs_read_prepare(struct rpc_task *task, void *calldata) | |
436 | { | |
437 | struct nfs_read_data *data = calldata; | |
438 | ||
035168ab | 439 | if (nfs4_setup_sequence(NFS_SERVER(data->inode), |
f11c88af AA |
440 | &data->args.seq_args, &data->res.seq_res, |
441 | 0, task)) | |
442 | return; | |
443 | rpc_call_start(task); | |
444 | } | |
445 | #endif /* CONFIG_NFS_V4_1 */ | |
446 | ||
ec06c096 | 447 | static const struct rpc_call_ops nfs_read_partial_ops = { |
f11c88af AA |
448 | #if defined(CONFIG_NFS_V4_1) |
449 | .rpc_call_prepare = nfs_read_prepare, | |
450 | #endif /* CONFIG_NFS_V4_1 */ | |
ec06c096 | 451 | .rpc_call_done = nfs_readpage_result_partial, |
fdd1e74c | 452 | .rpc_release = nfs_readpage_release_partial, |
ec06c096 TM |
453 | }; |
454 | ||
1de3fc12 TM |
455 | static void nfs_readpage_set_pages_uptodate(struct nfs_read_data *data) |
456 | { | |
457 | unsigned int count = data->res.count; | |
458 | unsigned int base = data->args.pgbase; | |
459 | struct page **pages; | |
460 | ||
79558f36 TM |
461 | if (data->res.eof) |
462 | count = data->args.count; | |
1de3fc12 TM |
463 | if (unlikely(count == 0)) |
464 | return; | |
465 | pages = &data->args.pages[base >> PAGE_CACHE_SHIFT]; | |
466 | base &= ~PAGE_CACHE_MASK; | |
467 | count += base; | |
468 | for (;count >= PAGE_CACHE_SIZE; count -= PAGE_CACHE_SIZE, pages++) | |
469 | SetPageUptodate(*pages); | |
0b671301 TM |
470 | if (count == 0) |
471 | return; | |
472 | /* Was this a short read? */ | |
473 | if (data->res.eof || data->res.count == data->args.count) | |
1de3fc12 TM |
474 | SetPageUptodate(*pages); |
475 | } | |
476 | ||
1da177e4 LT |
477 | /* |
478 | * This is the callback from RPC telling us whether a reply was | |
479 | * received or some error occurred (timeout or socket shutdown). | |
480 | */ | |
ec06c096 | 481 | static void nfs_readpage_result_full(struct rpc_task *task, void *calldata) |
1da177e4 | 482 | { |
ec06c096 | 483 | struct nfs_read_data *data = calldata; |
1da177e4 | 484 | |
0b671301 TM |
485 | if (nfs_readpage_result(task, data) != 0) |
486 | return; | |
fdd1e74c TM |
487 | if (task->tk_status < 0) |
488 | return; | |
1de3fc12 | 489 | /* |
0b671301 | 490 | * Note: nfs_readpage_retry may change the values of |
1de3fc12 | 491 | * data->args. In the multi-page case, we therefore need |
0b671301 TM |
492 | * to ensure that we call nfs_readpage_set_pages_uptodate() |
493 | * first. | |
1de3fc12 | 494 | */ |
fdd1e74c TM |
495 | nfs_readpage_truncate_uninitialised_page(data); |
496 | nfs_readpage_set_pages_uptodate(data); | |
497 | nfs_readpage_retry(task, data); | |
498 | } | |
499 | ||
500 | static void nfs_readpage_release_full(void *calldata) | |
501 | { | |
502 | struct nfs_read_data *data = calldata; | |
503 | ||
1da177e4 LT |
504 | while (!list_empty(&data->pages)) { |
505 | struct nfs_page *req = nfs_list_entry(data->pages.next); | |
1da177e4 | 506 | |
1de3fc12 | 507 | nfs_list_remove_request(req); |
1da177e4 LT |
508 | nfs_readpage_release(req); |
509 | } | |
fdd1e74c | 510 | nfs_readdata_release(calldata); |
1da177e4 LT |
511 | } |
512 | ||
ec06c096 | 513 | static const struct rpc_call_ops nfs_read_full_ops = { |
f11c88af AA |
514 | #if defined(CONFIG_NFS_V4_1) |
515 | .rpc_call_prepare = nfs_read_prepare, | |
516 | #endif /* CONFIG_NFS_V4_1 */ | |
ec06c096 | 517 | .rpc_call_done = nfs_readpage_result_full, |
fdd1e74c | 518 | .rpc_release = nfs_readpage_release_full, |
ec06c096 TM |
519 | }; |
520 | ||
1da177e4 LT |
521 | /* |
522 | * Read a page over NFS. | |
523 | * We read the page synchronously in the following case: | |
524 | * - The error flag is set for this page. This happens only when a | |
525 | * previous async read operation failed. | |
526 | */ | |
527 | int nfs_readpage(struct file *file, struct page *page) | |
528 | { | |
529 | struct nfs_open_context *ctx; | |
530 | struct inode *inode = page->mapping->host; | |
531 | int error; | |
532 | ||
533 | dprintk("NFS: nfs_readpage (%p %ld@%lu)\n", | |
534 | page, PAGE_CACHE_SIZE, page->index); | |
91d5b470 CL |
535 | nfs_inc_stats(inode, NFSIOS_VFSREADPAGE); |
536 | nfs_add_stats(inode, NFSIOS_READPAGES, 1); | |
537 | ||
1da177e4 LT |
538 | /* |
539 | * Try to flush any pending writes to the file.. | |
540 | * | |
541 | * NOTE! Because we own the page lock, there cannot | |
542 | * be any new pending writes generated at this point | |
543 | * for this page (other pages can be written to). | |
544 | */ | |
545 | error = nfs_wb_page(inode, page); | |
546 | if (error) | |
de05a0cc TM |
547 | goto out_unlock; |
548 | if (PageUptodate(page)) | |
549 | goto out_unlock; | |
1da177e4 | 550 | |
5f004cf2 TM |
551 | error = -ESTALE; |
552 | if (NFS_STALE(inode)) | |
de05a0cc | 553 | goto out_unlock; |
5f004cf2 | 554 | |
1da177e4 | 555 | if (file == NULL) { |
cf1308ff | 556 | error = -EBADF; |
d530838b | 557 | ctx = nfs_find_open_context(inode, NULL, FMODE_READ); |
1da177e4 | 558 | if (ctx == NULL) |
de05a0cc | 559 | goto out_unlock; |
1da177e4 | 560 | } else |
cd3758e3 | 561 | ctx = get_nfs_open_context(nfs_file_open_context(file)); |
1da177e4 | 562 | |
9a9fc1c0 DH |
563 | if (!IS_SYNC(inode)) { |
564 | error = nfs_readpage_from_fscache(ctx, inode, page); | |
565 | if (error == 0) | |
566 | goto out; | |
567 | } | |
568 | ||
8e0969f0 TM |
569 | error = nfs_readpage_async(ctx, inode, page); |
570 | ||
9a9fc1c0 | 571 | out: |
1da177e4 LT |
572 | put_nfs_open_context(ctx); |
573 | return error; | |
de05a0cc | 574 | out_unlock: |
1da177e4 LT |
575 | unlock_page(page); |
576 | return error; | |
577 | } | |
578 | ||
579 | struct nfs_readdesc { | |
8b09bee3 | 580 | struct nfs_pageio_descriptor *pgio; |
1da177e4 LT |
581 | struct nfs_open_context *ctx; |
582 | }; | |
583 | ||
584 | static int | |
585 | readpage_async_filler(void *data, struct page *page) | |
586 | { | |
587 | struct nfs_readdesc *desc = (struct nfs_readdesc *)data; | |
588 | struct inode *inode = page->mapping->host; | |
589 | struct nfs_page *new; | |
590 | unsigned int len; | |
de05a0cc TM |
591 | int error; |
592 | ||
49a70f27 | 593 | len = nfs_page_length(page); |
1da177e4 LT |
594 | if (len == 0) |
595 | return nfs_return_empty_page(page); | |
de05a0cc | 596 | |
1da177e4 | 597 | new = nfs_create_request(desc->ctx, inode, page, 0, len); |
de05a0cc TM |
598 | if (IS_ERR(new)) |
599 | goto out_error; | |
600 | ||
1da177e4 | 601 | if (len < PAGE_CACHE_SIZE) |
eebd2aa3 | 602 | zero_user_segment(page, len, PAGE_CACHE_SIZE); |
f8512ad0 FI |
603 | if (!nfs_pageio_add_request(desc->pgio, new)) { |
604 | error = desc->pgio->pg_error; | |
605 | goto out_unlock; | |
606 | } | |
1da177e4 | 607 | return 0; |
de05a0cc TM |
608 | out_error: |
609 | error = PTR_ERR(new); | |
610 | SetPageError(page); | |
611 | out_unlock: | |
612 | unlock_page(page); | |
613 | return error; | |
1da177e4 LT |
614 | } |
615 | ||
616 | int nfs_readpages(struct file *filp, struct address_space *mapping, | |
617 | struct list_head *pages, unsigned nr_pages) | |
618 | { | |
8b09bee3 | 619 | struct nfs_pageio_descriptor pgio; |
1da177e4 | 620 | struct nfs_readdesc desc = { |
8b09bee3 | 621 | .pgio = &pgio, |
1da177e4 LT |
622 | }; |
623 | struct inode *inode = mapping->host; | |
624 | struct nfs_server *server = NFS_SERVER(inode); | |
8b09bee3 TM |
625 | size_t rsize = server->rsize; |
626 | unsigned long npages; | |
5f004cf2 | 627 | int ret = -ESTALE; |
1da177e4 LT |
628 | |
629 | dprintk("NFS: nfs_readpages (%s/%Ld %d)\n", | |
630 | inode->i_sb->s_id, | |
631 | (long long)NFS_FILEID(inode), | |
632 | nr_pages); | |
91d5b470 | 633 | nfs_inc_stats(inode, NFSIOS_VFSREADPAGES); |
1da177e4 | 634 | |
5f004cf2 TM |
635 | if (NFS_STALE(inode)) |
636 | goto out; | |
637 | ||
1da177e4 | 638 | if (filp == NULL) { |
d530838b | 639 | desc.ctx = nfs_find_open_context(inode, NULL, FMODE_READ); |
1da177e4 LT |
640 | if (desc.ctx == NULL) |
641 | return -EBADF; | |
642 | } else | |
cd3758e3 | 643 | desc.ctx = get_nfs_open_context(nfs_file_open_context(filp)); |
9a9fc1c0 DH |
644 | |
645 | /* attempt to read as many of the pages as possible from the cache | |
646 | * - this returns -ENOBUFS immediately if the cookie is negative | |
647 | */ | |
648 | ret = nfs_readpages_from_fscache(desc.ctx, inode, mapping, | |
649 | pages, &nr_pages); | |
650 | if (ret == 0) | |
651 | goto read_complete; /* all pages were read */ | |
652 | ||
94ad1c80 | 653 | pnfs_pageio_init_read(&pgio, inode); |
8b09bee3 TM |
654 | if (rsize < PAGE_CACHE_SIZE) |
655 | nfs_pageio_init(&pgio, inode, nfs_pagein_multi, rsize, 0); | |
656 | else | |
657 | nfs_pageio_init(&pgio, inode, nfs_pagein_one, rsize, 0); | |
658 | ||
1da177e4 | 659 | ret = read_cache_pages(mapping, pages, readpage_async_filler, &desc); |
8b09bee3 TM |
660 | |
661 | nfs_pageio_complete(&pgio); | |
662 | npages = (pgio.pg_bytes_written + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; | |
663 | nfs_add_stats(inode, NFSIOS_READPAGES, npages); | |
9a9fc1c0 | 664 | read_complete: |
1da177e4 | 665 | put_nfs_open_context(desc.ctx); |
5f004cf2 | 666 | out: |
1da177e4 LT |
667 | return ret; |
668 | } | |
669 | ||
f7b422b1 | 670 | int __init nfs_init_readpagecache(void) |
1da177e4 LT |
671 | { |
672 | nfs_rdata_cachep = kmem_cache_create("nfs_read_data", | |
673 | sizeof(struct nfs_read_data), | |
674 | 0, SLAB_HWCACHE_ALIGN, | |
20c2df83 | 675 | NULL); |
1da177e4 LT |
676 | if (nfs_rdata_cachep == NULL) |
677 | return -ENOMEM; | |
678 | ||
93d2341c MD |
679 | nfs_rdata_mempool = mempool_create_slab_pool(MIN_POOL_READ, |
680 | nfs_rdata_cachep); | |
1da177e4 LT |
681 | if (nfs_rdata_mempool == NULL) |
682 | return -ENOMEM; | |
683 | ||
684 | return 0; | |
685 | } | |
686 | ||
266bee88 | 687 | void nfs_destroy_readpagecache(void) |
1da177e4 LT |
688 | { |
689 | mempool_destroy(nfs_rdata_mempool); | |
1a1d92c1 | 690 | kmem_cache_destroy(nfs_rdata_cachep); |
1da177e4 | 691 | } |